INTRODUCTION TO GENERATIVE LEXICON
INTRODUCTION TO GENERATIVE LEXICON James Pustejovsky (2008) Lexical Semantics 12/17 OUTLINE • Introduction • Traditional Lexical Representations • The Nature of Polysemy • Levels of Lexical Meaning • Coercion and Compositionality • Complex Types in Language • Conclusion INTRODUCTION • Generative Lexicon (GL) attempts to spread the semantic load across all constituents of the utterance. • Two major lines of inquiry: 1. How is it that we are able to deploy a finite number of words in our language in an unbounded number of contexts? 2. Is lexical information and the representations used in composing meanings separable from our commonsense knowledge? INTRODUCTION • Generative Lexicon introduces a knowledge representation framework which offers a rich and expressive vocabulary for lexical information. • Lexicon may evolved contrast to currently prevalent views of static lexicon design. FOUR LEVELS • The computational resources available to a lexical item consist of the following four levels: 1. Lexical typing structure: giving an explicit type for a word positioned within a type system for the language 2. Argument structure: specifying the number and nature of the arguments to a predicate 3. Event structure: defining the event type of the expression and any subeventual structure it may have 4. Qualia structure: this defines the essential attributes of objects, events, and relations, associated with a lexical item A BRIEF INTRO TO QS • The qualia structure, inspired by Moravcsik’s (1975) interpretation of the aitia (cause) of Aristotle. • The modes of explanation associated with a word or phrase in the language: 1. Formal: the basic category of which distinguishes the meaning of a word within a larger domain. 2. Constitutive: the relation between an object and its constituent parts. 3. Telic: the purpose or function of the object, if there is one. 4. Agentive: the factors involved in the object’s origins or “coming into being”. A BRIEF INTRO TO QS TRADITIONAL LEXICAL REPRESENTATION • Word meaning can be exhaustively defined by an enumerable set of senses per word. • The problem of lexical ambiguity: • to select the most appropriate ‘definition’ available • the selection process is driven by matching sense characterizations against contextual factors • fails to account for the creative use of words in novel contexts (e.g. water the plant; water the whisky) GL ACCOUNT FOR AMBIGUITY • Extensibility: it is open-ended in nature and accounts for the novel, creative, uses of words in a variety of contexts by positing procedures for generating semantic expressions for words on the basis of particular contexts. • Advantages: 1. a rich and expressive lexicon can explain aspects of learnability 2. offer improvements in robustness of coverage GL ACCOUNT FOR AMBIGUITY • The ambiguity problem is rephrased in terms of dynamic interpretation of a word in context. • Semantic expressions for word meaning in context are constructed by a fixed number of generative devices, which operate on a core set of senses. • Through composition, an extended set of word senses is obtained when individual lexical items are considered jointly with others in larger phrases. THE NATURE OF POLYSEMY • One of the most pervasive phenomena in natural language is that of systematic ambiguity, or polysemy. • Traditionally view is selecting for a particular word sense. • Problems • the numbers of and distinctions between senses within an entry are ‘frozen’ into a fixed grammar ’s lexicon. • definitions hardly make any provisions for the notion that boundaries between word senses may shift with context. EXAMPLE: FAST a. The island authorities sent out a fast little government boat to welcome us. b. a fast typist. c. Rackets is a fast game. d. a fast book. e. My friend is a fast driver and a constant worry to her cautious husband. • fast(1): moving quickly; fast(2): performing some act quickly; fast(3): doing something requiring a short space of time; fast(4): involving rapid motion. EXAMPLE: FAST • In fact, any finite enumeration of word senses will not account for creative applications of this adjective in the language. • E.g. fast motorway and fast garage • As novel uses of fast, we are clearly looking at new senses which are not covered by the enumeration given above. EXAMPLE: BAKE • If we assume that enumeration is adequate as a descriptive mechanism, it is not always obvious how to select the correct word sense in any given context: a. John baked the potatoes. b. Mary baked a cake. • The problem here is that there is too much overlap in the ‘core’ semantic components of the different readings. LEVELS OF LEXICAL MEANING • A lexical entry for a word encodes a range of representative aspects of lexical meaning. • When embedded in the context of other words, mutually compatible roles in the lexical decompositions of each word become more prominent. Context A A1 A2 … Lexical C C1 C2 (…) Context B B1 B2 … Lexical C C3 C4 (…) LEVELS OF LEXICAL MEANING • All lexical types play in contributing to the overall meaning of a phrase. • compositionality, necessarily different from the more conventional pairing of verbs as functions and nouns as arguments. • spreading the semantic load evenly across the lexicon. • The ‘static’ definition of a word provides its literal meaning; it is only through the suitable composition of appropriately highlighted projections of words that we generate new meanings in context. (e.g. water the plant; water the whisky) FOUR LEVELS OF REPRESENTATIONS • Lexical Typing Structure • This determines the ways in which a word is related to other words in a structured type system • link to general world (commonsense) knowledge • Argument Structure • specifying the number and nature of the arguments to a predicate • Event Structure • this identifies the particular event type for a verb or a phrase • the primitive event type—state (S), process (P) or transition (T) • the focus of the event • Qualia structure QUALIA STRUCTURE • Qualia structure is a system of relations that characterizes the semantics of a lexical item. • The semantics of nominal will be the focus here. • The elements that make up a qualia structure include familiar notions such as container, space, surface, figure, or artifact. • The qualia structure along with the other representational devices (event structure and argument structure) can be seen as providing the building blocks for possible object types. QUALIA STRUCTURE • The modes of explanation associated with a word or phrase in the language: 1. FORMAL: the basic category of which distinguishes the meaning of a word within a larger domain. 2. CONSTITUTIVE: the relation between an object and its constituent parts. 3. TELIC: the purpose or function of the object, if there is one. 4. AGENTIVE: the factors involved in the object’s origins or “coming into being”. QUALIA STRUCTURE • This figure illustrates a type hierarchy fragment for knowledge about objects, encoding qualia structure information. • The tangled type hierarchy above shows how qualia can be unified to create more complex concepts out of simple ones. QUALIA STRUCTURE • We can distinguish the domain of individuals into three ranks or levels of type: 1. Nature types: Natural kind concepts consisting of reference only to Formal and Const qualia roles; 2. Functional types: Concepts integrating reference to purpose or function. 3. Complex types: Concepts integrating reference to a relation between types. QUALIA STRUCTURE be given a function (i.e, a Telic role) • Nature type • Functional type • The noun sandwich contains information of the “eating activity” as a constraint on its Telic value. • Mary finished her sandwich. Mary eating the sandwich. • Type coercion COERCION AND COMPOSITIONALITY • Type Coercion: A semantic operation that converts an argument to the type which is expected by a function (predicate), where it would otherwise result in a type error. • By allowing lexical items to coerce their arguments, we obviate the enumeration of multiple entries for different senses of a word. EXAMPLE: EN JOY AND B EG IN • The notion that a predicate can specify a particular target type for its argument is a very useful one, and intuitively explains the different syntactic argument forms: 1. a. Mary enjoyed the movie. b. Mary enjoyed watching the movie. 2. a. Mary began a book. b. Mary began reading a book. c. Mary began to read a book. • Noun phrases and verb phrases appear in the same argument position, somehow satisfying the type required by the verbs enjoy and begin. EXAMPLE: KILL AND WA KE • Noun phrases of very different semantic classes appear as subject of the verbs kill and wake: 1. a. John killed Mary. b. The gun killed Mary. c. The bullet killed Mary. 2. a. The cup of coffee woke John up. b. Mary woke John up. c. John’s drinking the cup of coffee woke him up. • The verbs seem to be interpreting all the phrases as events of some sort. • We are unable to capture the underlying relatedness between these entries if just following the sense enumeration theory. COMPLEX TYPES IN LANGUAGE • One of the more unique aspects of the representational mechanisms of GL is the data structure known as a complex type (or dot object). • A single word may refer to two or more aspects of an object’s meaning or concept. • window, door, fireplace, and room can be used to refer to (1)the physical object itself or (2)the space associated with it: 1. a. They walked through the door. b. She will paint the door red. COMPLEX TYPES IN LANGUAGE • As with nouns such as door, the nouns book and exam denote two contradictory types; books are both physical form and informational in nature; exams are both events and informational: 1. a. Mary doesn’t believe the book. b. John bought his book from Mary. c. The police burnt a controversial book. COMPLEX TYPES IN LANGUAGE • There must exist a relation R which relates the elements of the pairing. • For nouns such as book, disk, and record, the relation R is a species of “containment,” and shares grammatical behavior with other container-like concepts. information in a book The lexical structure for book as a dot object CONCLUSION • The challenges posed by Generative Lexicon to linguistic theory: semantic interpretation is as creative and generative as syntax. • It is GL’s goal to uncover these mechanisms in order to model the expressive semantic power of language.
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project
Technical Report The acquisition of a unification-based generalised categorial grammar Aline Villavicencio
Semi-Compositional Noun + Verb Constructions: Theoretical Questions and Computational Linguistic Analyses